Tintri and VMware showed the virtualisation players how it's done at VMWorld today when they tested a Tintri hybrid storage array booting 1,000 virtual desktops in two-and-a-half hours – nine seconds per desktop. Tintri hybrid flash and disk drive arrays are purpose-designed to support virtual server operations. The test set up …
Just trying to figure out how this is news worthy. NetApp with Flexclone and Flash Cache have been able to do this for a while. Am I missing something?
"2.5 hours, 150 minutes"
Is that 5 hours then? Or 300 minutes?
Although it's an easy equation, booting 1000 desktops in 2.5 hrs doesn't mean they booted in 9 secs - if you had 1000 users all trying to 'boot' at 9.00 and some weren't ready until 11.30 it's still 2.5 hours until everyone is productive..
Faster is Better
Two and a half hours is nice. 5 minutes would be better in an all-flash scenario.
I think they actually mean 'provisioned' not 'booted' in 150mins.
how large were the images and were they thin or thick provisioned. these factors are an important metric in deciding is this is genuine speed, or just another massively favourable sales pitch.
Boot time or setup time?
I really hope they mean 2.5 hours to create the thousand instances - a 2.5 hour boot time would be virtually worthless! Equally, spreading the boot times over 2.5 hours isn't much use for VDI; surely most real VDIs would be powering up very close to the 9am mark, very few earlier than 8:30 or later than 9:30.
Having said that, I recall Novell bragging about having built a big (100k user?) authentication setup, back in the 90s. I was horrified by the two-minute login times they quoted. Guess what we ended up having rolled out to us. (Result: workflow consisting of 'enter username & password, go make coffee, come back to logged in desktop you hope'.) Quite why it couldn't process logins at a more adequate pace, I don't know.
Merely booting a thousand machines should entail something of the order of a terabyte of network traffic; saturating a single 10GE link, that should be of the order of 20 minutes, not 2.5 hours (for a single cloned OS image, you should have about 1 Gb of 'hot' data, being served from RAM for all but the first boot). Now, show it having all the users starting to come in at 8:50 and being booted and ready to go by 9:10 I'll be more impressed - but spread it out over enough hours, you could boot them all from one Mac Mini with some SATA drives, it just won't be fast enough to be useful!
Seems very slow
2.5 hours for 1000 VDI desktop would make for very interesting Monday mornings at a big corporation. It just isn't very fast.
This may be a measure of some sort for raw ooomph, but I've helped create much faster demos. One of them, using a single NetApp filer with a flash accelerator card, managed 8000 desktops in less than 10 minutes. This is what you would expect with two 10GE links, which is where the bottleneck occurs.
At $180,000, this is not a good way to solve the boot storm problem. VMWare and Tintri need to go back to the drawing board and look at network topology and test conditions very closely, as well as the caching policy in the array.
The article does a poor job of describing the technology or the approach used.
First, the Tintri provides much higher front-end IOPS (especially write IOPS) than an equivalently-priced NetApp or EMC product. It also provides much more capacity than an all-Flash solution at that price (Pure Storage *might* be able to deliver equivalent capacity, depending on deduplication and compression, but it's hard to know up front).
Second, this is not just about *booting* the virtual desktops but about *provisioning* them from scratch. Booting 1000 existing desktops ought to be much quicker.
Full disclosure: I'm speaking as a satisfied Tintri customer who has migrated a significant quantity of desktop and server VMs off of NetApp with Flash Cache and hasn't looked back.
Perhaps VMWare and Tintri could issue an explanation of what they actually are boasting about!
I suspect Tintri and VMware conveyed the explanation just fine but that the finer points were lost in translation.
The test was a complete provisioning, booting and application usage test for 1,000 desktops
To clarify the points raised by readers, the test was a complete provisioning, booting and application usage test for 1,000 desktops – all at “Macbook Air” performance levels, so the 150 minutes refers to the time taken for 1,000 linked clones to be fully deployed and the tests completed on a single Tintri VMstore T540 storage node.
An overview of the testing can be viewed here - http://www.tintri.com/wp-content/uploads/2012/10/Reexamining-VDI-Conventions-121008.pdf
Disclosure: I am VP of global marketing at Tintri and quoted within the article